Coherence Implementation Patterns - Sig Nov 2011
•
8 likes•7,084 views
Notes from the Nov Coherence Sig
Recommended
Recommended
More Related Content
What's hot
What's hot (20)
Viewers also liked
Viewers also liked (11)
Similar to Coherence Implementation Patterns - Sig Nov 2011
Similar to Coherence Implementation Patterns - Sig Nov 2011 (20)
More from Ben Stopford
More from Ben Stopford (18)
Recently uploaded
Recently uploaded (20)
Coherence Implementation Patterns - Sig Nov 2011
- 1. Coherence Implementa/on Pa1erns Ben Stopford The Royal Bank of Scotland
- 2. Some Ideas Nothing More
- 3. Why do we use Coherence? Fast? Scalable? Applica/on layer?
- 5. • We don’t want ACID all of the /me • We want to pick the bits we need when we need them • We want to use the context of our business requirement to work our way around the ones we don’t need.
- 7. Why do we care? Without versioning it’s a free-‐for-‐all. • What changed? • Was something overwri1en? • How can you prevent concurrent updates? • What did the system look like 10 seconds ago. • How can I provide a consistent view? • How to I ensure ordering of updates in an asynchronous system?
- 8. Versioning your Objects Versioned Cache A.1 A.2 B.1 C.1 C.2 D.1
- 9. Versioning your Objects Cache A.1 Coherence Trigger A.2 New Version = Old Version + 1 ??
- 10. Running a Coherence Filter
- 12. Latest / Versioned Pa1ern Coherence Trigger Write Versioned Latest Cache Cache A A.1 B A.2 C B.1 D C.1 C.2 D.1
- 13. Latest / Versioned Pa1ern Latest Cache Key = [Business Key] Versioned Cache Key = [Business Key][Version] !" !&'" #" !&(" $" #&'" %" $&'" $&(" %&'"
- 14. Suffers from data duplica/on
- 15. Latest Marker Pa1ern Write Versioned Coherence Trigger Cache A.1 A.L B.L Well Known Marker Version C.1 C.L D.L
- 16. However our trigger can’t use cache.put() Why?
- 17. Need to consider the threading model
- 18. So we’ll need to use the backing map directly public void copyObjectToVersionedCacheAddingVersion(MapTrigger.Entry entry) { MyValue value = (MyValue)entry.getValue(); MyKey versionedKey = (MyKey)value.getKey(); BinaryEntry binary = (BinaryEntry)entry; Binary binaryValue = binaryEntry.getBinaryValue(); Map map = binary.getContext().getBackingMap("VersionedCacheName"); map.put(toBinary(versionedKey), binaryValue); }
- 19. A third approach
- 20. The Collec/ons Cache Trigger Appends to Collec/on collec/onsCache.put(key, val); A [O1, O2, O3..] B [O1, O2, O3..] collec/onsCache.invoke(key, C [O1, O2, O3..] new LastValueGe1er()); …or override backing store D [O1, O2, O3..] Collec/onsCache
- 21. So we have 3 pa1erns for managing versioning whilst retaining key based access
- 22. Using versioning to manage concurrent changes Mul/ Version Concurrency Control (MVCC)
- 23. Cache Version 1 Coherence Trigger Version 2 New Version = Old Version + 1 ??
- 24. Concurrent Object Update (2 Clients update the same object at the same /me) Client1 Client2 A.1 A.2 A.2
- 25. Concurrent Object Update (Client2 fails to update dirty object) Client1 Client2 A.1 A.2 Versioned Cache
- 26. Concurrent Object Update (Client 2 updates clean object) Client1 Client2 A.1 A.2 A.3 Versioned Cache
- 27. So a concurrent update results in an error and must be retried. &'()*+#$ &'()*+%$ !"#$ !"%$
- 28. What’s going to happen if we are using putAll?
- 29. Reliable PutAll We want putAll to tell us which objects failed the write process
- 30. Reliable PutAll Node Node Client Extend Node Node Node Node Invocable: • Split keys by member Invocable: • Send appropriate values • Write entries to to each member backing map (we • Collect any excep/ons use an EP for returned this)
- 31. This gives us a reliable mechanism for knowing what worked and what failed
- 33. The Fat Object Method Cache A B C D
- 34. The Single Entry Point Method (objects are stored separately) Collocate with key associa/on
- 35. All writes synchronize on the primary object. EP
- 36. All reads synchronize on the primary object. EP
- 37. Wri/ng Orphaned Objects Write read entry point object last Write orphaned objects first This mechanism is subtly flawed
- 38. Reading several objects as an atomic unit aka Joins
- 39. The trivial approach to joins Get Get Get Get Get Get Get Cost Ledger Source Transac MTMs Legs Cost Centers Books Books -‐/ons Centers Network Time
- 40. Server Side, Sharded Joins Use KeyAssocia/on to keep related en//es together Orders Shipping Log Common Key
- 41. Server Side, Sharded Joins Transactions Aggregator joins Mtms data across cluster Cashflows
- 42. So we have a set of mechanisms for reading and wri/ng groups of related objects.
- 44. A service that automa/cally restarts amer failure
- 45. A service that automa/cally restarts amer failure
- 46. What is the cluster singleton good for • Adding indexes • Loading data • Keeping data up to date • Upda/ng cluster /me • You can probably think of a bunch of others yourselves.
- 47. Code for Cluster Singleton //run in a new thread on every Cache Server while (true) { boolean gotLock = lockCache.lock("singletonLock", -1); if (gotLock) { //Start singletons wait(); } }
- 48. Implemen/ng Consistent Views and Repeatable Queries
- 49. Bi-‐temporal public interface MyBusinessObject{ //data Business public Date getBusinessDate(); Time public Date validFrom(); System public Date validTo(); Time }
- 50. Where does the System Time come from?
- 51. You can’t use the System.currentTimeMillis() in a distributed environment!
- 52. You need a cluster synchronised clock
- 53. Repeatable Time: A guaranteed Tick Write first Write Time Singleton Service Read Time Write second Replicated Caches (pessimis/c)
- 54. As we add objects we /mestamp them with Write Time Write Time Singleton Service
- 55. When we read objects we use Read Time Singleton Service Read Time
- 56. Repeatable Time: A guaranteed Tick 8 7 8 7 8 7 Write first Write Time Singleton Service Read Time Write second 6 7 6 7 6 7 Replicated Caches (pessimis/c)
- 57. Event Based Processing Async Cachestore cache.put(key, val); A B C D
- 58. Messaging as a System of Record Messaging System (use Topics for scalability)
- 59. Messaging as a System of Record Trigger cache.put(key, val); A B JMS C TOPIC D
- 60. Easy Grid Implementa/on in GUIs
- 61. CQCs on a CQC GUI JVM CQC CQC Cache Define this in config
- 62. How do you release quickly to a Coherence cluster?
- 63. Rolling Restart?
- 64. Disk-‐Persist
- 65. Final Thoughts
- 66. Data is the most important commodity that you have Keep it safe
- 67. Use a Par//on Listener
- 68. Have Proac/ve Monitoring of Memory
- 70. Thanks Slides & related ar/cles available at: h1p://www.benstopford.com